Shared cache meeting 20200624

Present at the meeting:

• Arseniy Alekseyev (@aalekseyev)
• Andrey Mokhov (@snowleopard)
• Jérémie Dimino (@jeremiedimino)
• Quentin Hocquet (@mefyl)

Discussion

• The distributed cache HTTP protocol implementation is working, client can connect to it and send queries
• Currently one backend server, need to design a configuration file format to connect to multiple servers
• Still need to figure out how each server finds itself in the configuration file. Agreed that the hostname should do fine for self-identification.
• As for specification of hash assignment: we can have just a list of machines and hash ranges assigned to each. Using a prefix of a hash could be a shorthand syntax for specifying a range.
• Not much else is remaining to do on the Quentin side.
• Andrey was testing and cleaning up the new cache format in jenga
• We discussed access permission management for the future versions. For now the idea is to have the reverse proxy do authentication and forward the username to the cache daemon, but it's the cache daemon itself who will be responsible for authorization by using the "repo" field in the metadata and knowing which users are allowed to access which repos. This has an undesirable consequence of the daemon having to understand metadata, but that should be fine. In any case, none of this is going to be there in the first version. (although we already have commit and repo names in the metadata, even if they are unused)

Next steps

• On the Quentin's side, the next step is to implement the config file and the ability to connect to multiple servers (~1 week ETA)
• On the Jane Street side, We need to write a client-side implementation of the protocol between jenga and the cache daemon and then we can test inside Jane Street (~2 weeks ETA)